Optimizing stimulus energy for cochlear implants with a machine learning model of the auditory nerve.

Area of Science:

• Computational neuroscience
• Biophysics
• Machine learning applications

Background:

• Simulations of realistic biophysical auditory nerve fiber models are computationally intensive.
• Efficient modeling is crucial for advancing auditory research and prosthetics.

Purpose of the Study:

• To develop an efficient surrogate model for auditory nerve fiber simulations using machine learning.
• To optimize stimulus waveform shape for energy efficiency using the surrogate model.

Main Methods:

• Developed a machine learning surrogate model, comparing various algorithms.
• Utilized a Convolutional Neural Network (CNN) for its superior performance.
• Implemented an Evolutionary Algorithm with the CNN surrogate to optimize stimulus waveforms.

Main Results:

• The CNN surrogate model achieved high accuracy (R²>0.99) in emulating the auditory nerve fiber model.
• Simulation time was reduced by five orders of magnitude.
• Optimized waveforms showed 8%-45% energy reduction compared to square waves.

Conclusions:

• The developed CNN surrogate model serves as an accurate and efficient replacement for complex auditory nerve fiber simulations.
• Optimized stimulus waveforms offer significant energy savings, beneficial for neural stimulation applications.